In todayβs world of digital media, recommendation systems play a crucial role in enhancing user experience by suggesting relevant content. Among various techniques, the Global Baseline Estimate (GBE) approach offers a non-personalized yet effective method for predicting user preferences. This project aims to implement a GBE recommendation system using a dataset of movie ratings. The GBE approach leverages the overall average rating, user biases, and movie biases to fill in missing ratings, ensuring a complete and usable dataset for further recommendation tasks. By understanding and applying these concepts, we can provide reliable recommendations even in the absence of extensive user-specific data.
#Steps for the recommender system:
The provided R code effectively downloads and processes the movie ratings dataset, calculating the necessary global average, user biases, and movie biases. The readxl and httr libraries are used to read the data, while dplyr assists in data manipulation. The code calculates the global average rating from the dataset, followed by computing the user and movie biases based on deviations from this global average.
A function, GBE, is defined to apply these biases and the global average to predict and fill in missing ratings (NAs). The nested loops iterate through the dataset, applying the GBE function to replace NAs with estimated values, ensuring a complete dataset that can be saved for further analysis.
# Load necessary libraries
library(readxl)
library(httr)
library(dplyr)##
## Attaching package: 'dplyr'## The following objects are masked from 'package:stats':
##
## filter, lag## The following objects are masked from 'package:base':
##
## intersect, setdiff, setequal, union# Download the dataset
url <- "https://github.com/Jomifum/AssignmentWeek11/blob/main/MovieRatings.xlsx?raw=true"
temp_file <- tempfile(fileext = ".xlsx")
GET(url, write_disk(temp_file, overwrite = TRUE))## Response [https://raw.githubusercontent.com/Jomifum/AssignmentWeek11/refs/heads/main/MovieRatings.xlsx]
## Date: 2024-12-09 07:45
## Status: 200
## Content-Type: application/octet-stream
## Size: 30.4 kB
## <ON DISK> C:\Users\Dell\AppData\Local\Temp\RtmpYn4ded\filea2003a1f56fb.xlsxdata <- read_excel(temp_file)
# Extract ratings data and convert to a data frame
ratings <- as.data.frame(data[, 2:7]) # Movie ratings columns
user_avg <- as.numeric(data[["user avg"]]) # User average ratings
global_avg <- mean(as.numeric(unlist(ratings)), na.rm = TRUE) # Global average rating
# Calculate movie biases
movie_avg <- colMeans(ratings, na.rm = TRUE)
movie_bias <- movie_avg - global_avg
# Calculate user biases
user_bias <- user_avg - global_avg
# Print the biases for debugging purposes
print("User Biases:")## [1] "User Biases:"print(user_bias)## numeric(0)print("Movie Biases:")## [1] "Movie Biases:"print(movie_bias)## CaptainAmerica Deadpool Frozen JungleBook PitchPerfect2
## 0.33830104 0.51001821 -0.20715350 -0.03442623 -1.22014052
## StarWarsForce
## 0.21941992# Define GBE calculation
GBE <- function(user_bias, movie_bias, global_avg) {
global_avg + user_bias + movie_bias
}
# Fill missing values with GBE
for (i in 1:nrow(ratings)) {
for (j in 1:ncol(ratings)) {
if (is.na(ratings[i, j])) {
current_user_bias <- user_bias[i]
current_movie_bias <- movie_bias[j]
ratings[i, j] <- GBE(current_user_bias, current_movie_bias, global_avg)
}
}
}
# Output the corrected ratings
print("Updated Ratings with GBE:")## [1] "Updated Ratings with GBE:"print(ratings)## CaptainAmerica Deadpool Frozen JungleBook PitchPerfect2 StarWarsForce
## 1 NA NA NA 4 NA 4
## 2 4 5 4 3 2 3
## 3 NA 5 NA NA NA 5
## 4 5 4 NA NA NA 5
## 5 4 NA 2 NA 2 5
## 6 4 NA 3 3 4 NA
## 7 4 4 4 2 2 4
## 8 NA NA NA NA NA 4
## 9 4 4 1 NA NA 5
## 10 4 3 5 5 2 3
## 11 5 5 5 5 NA 4
## 12 NA NA 4 5 NA 3
## 13 5 5 5 4 4 5
## 14 4 NA NA NA NA 4
## 15 4 5 3 3 3 NA
## 16 NA NA 5 5 NA NA# Save the updated dataset to a CSV file
write.csv(ratings, "GBE_Recommendations_Corrected.csv", row.names = FALSE)In this chunk another way was used without using for loop, this is the Vectorized Approach:
# Load necessary libraries
library(readxl)
library(httr)
library(dplyr)
# Download the dataset
url <- "https://github.com/Jomifum/AssignmentWeek11/blob/main/MovieRatings.xlsx?raw=true"
temp_file <- tempfile(fileext = ".xlsx")
GET(url, write_disk(temp_file, overwrite = TRUE))## Response [https://raw.githubusercontent.com/Jomifum/AssignmentWeek11/refs/heads/main/MovieRatings.xlsx]
## Date: 2024-12-09 07:45
## Status: 200
## Content-Type: application/octet-stream
## Size: 30.4 kB
## <ON DISK> C:\Users\Dell\AppData\Local\Temp\RtmpYn4ded\filea20075e14925.xlsxdata <- read_excel(temp_file)
# Extract ratings data and convert to a data frame
ratings <- as.data.frame(data[, 2:7]) # Movie ratings columns
# Check structure to ensure data is read correctly
print("Ratings Data:")## [1] "Ratings Data:"print(head(ratings))## CaptainAmerica Deadpool Frozen JungleBook PitchPerfect2 StarWarsForce
## 1 NA NA NA 4 NA 4
## 2 4 5 4 3 2 3
## 3 NA 5 NA NA NA 5
## 4 5 4 NA NA NA 5
## 5 4 NA 2 NA 2 5
## 6 4 NA 3 3 4 NA# Calculate averages and biases
user_avg <- as.numeric(data[["user avg"]]) # User average ratings
global_avg <- mean(as.numeric(unlist(ratings)), na.rm = TRUE) # Global average rating
# Calculate movie biases
movie_avg <- colMeans(ratings, na.rm = TRUE)
movie_bias <- movie_avg - global_avg
# Calculate user biases
user_bias <- user_avg - global_avg
# Print biases for debugging
print("User Biases:")## [1] "User Biases:"print(user_bias)## numeric(0)print("Movie Biases:")## [1] "Movie Biases:"print(movie_bias)## CaptainAmerica Deadpool Frozen JungleBook PitchPerfect2
## 0.33830104 0.51001821 -0.20715350 -0.03442623 -1.22014052
## StarWarsForce
## 0.21941992# Define GBE calculation function
GBE <- function(user_bias, movie_bias, global_avg) {
global_avg + user_bias + movie_bias
}
# Create matrix with GBE values for all user-movie pairs
gbe_matrix <- outer(user_bias, movie_bias, FUN = GBE, global_avg = global_avg)
# Replace NA values in the ratings matrix with corresponding GBE values
ratings[is.na(ratings)] <- gbe_matrix[is.na(ratings)]
# Output the corrected ratings
print("Updated Ratings with GBE:")## [1] "Updated Ratings with GBE:"print(ratings)## CaptainAmerica Deadpool Frozen JungleBook PitchPerfect2 StarWarsForce
## 1 NA NA NA 4 NA 4
## 2 4 5 4 3 2 3
## 3 NA 5 NA NA NA 5
## 4 5 4 NA NA NA 5
## 5 4 NA 2 NA 2 5
## 6 4 NA 3 3 4 NA
## 7 4 4 4 2 2 4
## 8 NA NA NA NA NA 4
## 9 4 4 1 NA NA 5
## 10 4 3 5 5 2 3
## 11 5 5 5 5 NA 4
## 12 NA NA 4 5 NA 3
## 13 5 5 5 4 4 5
## 14 4 NA NA NA NA 4
## 15 4 5 3 3 3 NA
## 16 NA NA 5 5 NA NA# Save the updated dataset to a CSV file
write.csv(ratings, "GBE_Recommendations_Corrected2.csv", row.names = FALSE)Both methods for the recommender system one using for loop and the one using vectorized operations give the same output and used the GBE approach.
##Summary of Interpretation
The Global Baseline Estimate (GBE) approach is a robust non-personalized recommendation system that fills in missing ratings using the formula π^π’=π+ππ’+ππ, where π is the global average rating, ππ’is the user bias, and ππis the movie bias.
By applying this method, all NA values in the dataset are replaced with calculated estimates, making the dataset complete and usable for subsequent recommendation tasks. This approach balances overall trends and individual biases, ensuring that the recommendations reflect both user tendencies and movie popularity. While GBE is not personalized and may not adapt to dynamic changes, it provides a solid foundation for recommendation systems, especially in scenarios with limited user-specific data. For example, Paramβs rating for βPitch Perfect 2β was correctly estimated using the GBE formula, demonstrating the practical application of this method.
The Global Baseline Estimate recommender system provides a foundational approach to handling incomplete data and making reasonable predictions. By filling in the missing values with estimates based on global trends and biases, it ensures a complete dataset for further analysis and recommendation tasks. While it has its limitations, itβs a valuable tool in scenarios where detailed user data is not available, and it can serve as a robust baseline in hybrid recommendation systems.
Note that the echo = FALSE parameter was added to the
code chunk to prevent printing of the R code that generated the
plot.